Topic 5: Misspecifications Flashcards
Omitted Relevant Variable
- what are the consequences?
Biased coefficient estimates of the included variables correlated with the omitted ones
Omitted Relevant Variable
- what is it and what causes it?
Variable that is correlated with the included variables but not included in the model
Omitted Relevant Variable
- how to detect it?
- theory
- significant unexpected signs
- RESET test
Omitted Relevant Variable
- remedy?
Include omitted variable or a proxy
Irrelevant Variable
- what is it and what causes it?
The inclusion of an unnecessary variable
Irrelevant Variable
- what are the consequences?
Lowers precision of model
• inflated standard errors
• low t-ratios
Irrelevant Variable
- how to detect it?
- theory
- t-test on beta
- adjusted r^2 increases if variable is dropped
Irrelevant Variable
- remedy?
Exclude the irrelevant variable
Incorrect Functional Form
- what is it and what causes it?
The functional form of the model might not be linear
Incorrect Functional Form
- what are the consequences?
- biased and inconsistent estimates
* poor fit of model (low R^2)
Incorrect Functional Form
- how to detect it?
- theory
- Ramsey RESET
- scatter plot of Y with each of the X’s
Incorrect Functional Form
- remedy?
- transform data into logs to linearise model
* add higher order functions of the variables to capture curvature
Multicollinearity
- what is it and what causes it?
When some of the explanatory variables are highly correlated with one another
Multicollinearity
- what are the consequences?
- high R^2, coefficients high SEs -> low t-ratios
- regression sensitive to small changes
- wide confidence intervals for parameters, incorrect inferences from model
Multicollinearity
- how to detect it?
- Correlogram
* see R^2 of regression of X on all other X’s
Multicollinearity
- remedy?
- ignore if model is okay
- drop collinear variable
- transform correlated variables into a ratio
- collect more data (longer sample period, higher frequency obs)
Autocorrelation
- what is it and what causes it?
Observations of the residuals are correlated over time
Causes: • omitted variables/common shocks
• Business Cycle inertia
• Overlapping effect of shocks
• Model misspecification
Autocorrelation
- what are the consequences?
- unbiased but inefficient
- incorrect inferences
- inflated R^2
Autocorrelation
- how to detect it?
- Durbin Watson
- Breusch Godfrey
- Correlogram of residuals
- Ljung-box
Autocorrelation
- remedy?
- GLS (if form is known)
- Dynamic Models
- HAC coefficients
- SE Newey-West
Heteroscedasticity
- what is it and what causes it?
Variance of error term not constant for all observations Causes: • scale/size effects • measurement error • subpopulation differences • flow of info is time varying
Heteroscedasticity
- what are the consequences?
- unbiased but inefficient estimates
* end up drawing wrong conclusions from hypotheses testing because of incorrect standard errors
Heteroscedasticity
- how to detect it?
- visual inspection of residual plot graph
- white’s test
- engle’s LM test for ARCH
Heteroscedasticity
- remedy?
- GLS (if form is known)
- transform variables using logs
- white’s SE estimates
Seasonality
- what is it and what causes it?
Observations of the dependent variable are systematically higher/lower in certain periods Causes: • day of the week effect • January effect • Bank holiday effect • open/close market effect
Seasonality
- what are the consequences?
Serially correlated error
Seasonality
- how to detect it?
Dummy variable for the period where the pattern is observed
Seasonality
- remedy?
Intercept or slope dummy to account for seasonality
Normality
- what is it and what causes it?
When the residuals are not normally distributed
Causes: • outliers
• Heteroscedasticity
• Seasonality
Normality
- what are the consequences?
- test statistics do not follow normal distribution
- estimators not efficient
- SEs are biased leading to wrong inferences
Normality
- how to detect it?
- Bera-Jarque
- histogram of residuals
- skewness and jurros is
Normality
- remedy?
- dummy variable to knock out outliers
* GARCH model
Structural break
- what is it and what causes it?
Parameters are not constant over sample period
Structural break
- what are the consequences?
Biased coefficient estimates
Structural break
- how to detect it?
• chow test for structural break
Structural break
- remedy?
- split the period
* dummy variables to account for different behaviour over the periods
